Multiclassing in GLoG
It seems like multiclassing would work really well in GLoG. The whole system is designed for modularity and despite the variety of authors in the space, we've mostly agreed to a shared system of templates, features, and something shaped like an ability score.
In practice, GLoG classes have a certain teleological curve that creates issues with multiclassing. The first one or two templates provide features that establish resources and mechanics that define the class so that you get to start playing with your toys right away. Their final template tends to provide a "showstopper" - a signature ability, sometimes usable only once, with broad narrative impact. This is most problematic in features not gated by a resource. Features like extra attacks or automatic skill successes make for easily exploitable dips.
Multiclassing within a particular archetype also leads to unpleasant redundancies. A few months back, I had a specialist who wanted to become a FIGHTER after a particularly nasty interaction with a mind-controlling crow.
My specialist had the following:
Watchful: You are never surprised. In combat, you can take a full turn before anybody.
The FIGHTER has:
The Full Dark World: ...
When a peaceful situation escalates to violence, you are never caught by surprise. If you're the one to escalate it, you always catch your enemies by surprise.
The two are not only redundant, but parsing them simultaneously creates all sorts of lingering ambiguities like "do I always take my enemies by surprise?" and "do I get to take my full turn in addition to a surprise turn (if the GLoG we're using has such a thing)?"
We could chalk all this up to the inherently imbalanced nature of a game distributed across a thousand blogposts, but let's consider some potential systems for handling multiclassing.
"Quest For It"
In order to take a template from a new class, you must have a mentor, institution, or patron willing to share their knowledge.
The lowest rules overhead way to handle this would just be to require players to quest for access to a new class. This has a couple nice benefits. It gives the opportunity for a check-in or counter offer from the GM in the case where a player wants to do something exploit-y. It also ties the character's classes more closely to the specific characters, institutions, or magical powers in the game world.
I find this works best at lower player counts or at "closed" table games that meet regularly over long periods of time. On something like a West Marches server you quickly run into uneven interest and attention from GMs. You can end up waiting a long time to "quest for it" or end up questing for something that feels minor or inconsequential (and thus unsatisfying when you actually achieve it).
Hire a Tutor
In order to take a template from a new class, you must hire a tutor. In addition to any costs spent on leveling, pay an additional 1000 gold to cover lessons and materials.
A low interaction version of this that might be suitable for something like a high player count server game is requiring the players to pay a tutor. You could establish a flat rate - e.g. in an XP-for-gold game, the gold cost of first level. Pay that amount to a tutor or academy in order to gain a class template from a new class.
This opens up less story options, but requires much less adjudication. It acts something like applying a bonus XP-cost to multiclassing, but making it literal gold has the bonus effect of making a use for excess treasure and opening it up to loans and other means of acquiring quick cash with long term consequences.
No Doubling Back
Whenever you would take a template, take the next template in alphabetical order, regardless of your current class or past templates within a current class.
Now we get into weirder, more mechanical, potential systems. The going standard is that when you take a template from a new class, you take that classes first template. What if instead, you just took the next template in alphabetical order. For instance, you might be a Wizard A, Fighter B, Wizard C, Thief D.
Obviously this just introduces new opportunities for cheesy exploits, but to me at least there is something less aesthetically galling about a character snaking from career to career on their gradual journey to power, rather than constantly dipping back. This can be tricky if you miss a key resource feat (e.g. having Magic Dice in the first place), but I feel like the player interested in complicated multiclassing is unlikely to fall into that problem by accident.
Gestalt Characters
In addition to their class, a character may have a gestalt class. Whenever you take a template, you may declare that gestalt class. When you would gain features from your class, instead gain a number of features equal to the fewest offered by your class or gestalt class individually. Select those features from either class.
A throw-back to AD&D and maybe a better version of the previous suggestion. This will obviously get wild fast and is prone to all sorts of breakages, but it dodges the problem with redundant or incoherent selections of features or missing out on a key resources.
As a byproduct, this system enforces a limit on number of number of classes. I'm generally not a fan of this kind of hard bound, but it could be an option. I like it better in the context where you are doing something interesting or with classes to take advantage of the reduced combinatoric complexity and exploitability.
Feature Compression
Whenever you take a template from a new class, work back through your existing features with your GM to check for redundancies. For each redundant feature, merge them into a single feature that retains the full functionality of both.
This is the suggestion that is most ruling intensive, but also the one I ended up pushing for in my specialist/FIGHTER example. This has the effect that you cannot acquire redundant effects from new classes, but also critically opens up an opportunity for GM intervention. It also has the pleasant quality of making your classes descriptive rather than prescriptive - the features change just as you do.
Talk About It
Fittingly, if you adopt any of these systems, you'll likely adopt some hybrid of several. Fundamentally, multiclassing is a good thing. It means that players are engaged with the system enough to think about it mechanically as a mode of play that occurs away from the table. Open that up into a conversation where you (as either player or GM) are willing to compromise or adjust to help keep the game fun for you and your fellow players.