Yes, but there are difficulties. The 'walls' of this 'maze' are invisible, which is a big problem. I have a section in my 'classic dungeon' where there is a whole sub-level with invisible walls (they also shock you if you touch them, just to be even meaner). Nobody gets through that. Trying to navigate that stuff is just super nasty. There are almost no landmarks, it is hard to orient, even if you invent some tricks to (partially) map it. A mystery of this sort of similar. Nobody knows where any clue really leads. At best you must be VERY VERY specific and spell everything out in great detail. This was not so needed in Doyle's writing because most of the things described tied in to common knowledge people had. If Holmes found a clue, people could interpret it, at least to some degree. Of course, if the game is set in a modern 'real world' type setting, then this helps a lot here!
'sub-mysteries' I agree would give a sense of progress. So a mystery that was, perhaps, an intricate conspiracy, where it is revealed in layers, would be a good design. I feel like we're getting into some very elaborate game scenarios though, which are hard to actually implement.
I wasn't really talking about elaborate conspiracy, although I would use the same design scaled up to do that. I'm going to continue to use the metaphor of node based design, even though that's not actually what going on, precisely. I'll identify groups of places and faces that have obviously strong ties, either of proximity or relationship to form a 'sub-mystery'. Let me spin an example, which will illustrate better than some blathering.
1. I start with the meta connect that the Guild of Stevedores is transhipping illegal cargo for Duke Hufflepuff (who is either the next level or the big bad). Drugs go from a ship, to a warehouse, to a farm just outside of town, owned by the Duke's nephew (who's also in on it).
2. Then I build a bundle of faces and places around the guild, and make a list of the kinds of clues the PCs might find. Lets say you have the Guild Master, his secretary, and one particular group of dock workers. The PCs already have the name of a ship and Wharf inspector, which is what leads them here in the first place. There's the Guild Hall, a club the Master frequents, a couple of bars frequented by Dockers, plus the docks and warehouses.
3. Then I figure out what kind of information the PCs might find that connects the pieces. I decide the secretary is the go between for the Duke and his crew at the docks. I also decide that wagons pick up the cargo late at night and move it to the farm (late night could be noticed, plus maybe there are night watchmen). The crew connects to the ship and the warehouse and the crooked inspector. The Duke connects to the secretary and the farm and the ship (I just added that in because it makes sense). The Duke and the Guild Master connect to the club, as does the secretary. The Guild Master connects to the Inspector and the crew. All of these connections can potentially be identified through surveillance and interviews with locals.
4. Now for hard proof. There are the drugs, shipping manifests, building ownership, letters, bank drafts, diaries, notes, Ducal signets, club reservations, plus assorted other stuff that might be suggested by the NPC details. From that I'll produce a list of what might be found in what location. Keep in mind these aren't master criminals, nor are they going up against the CSI crew, so you can be liberal with clues.
I can do that whole nexus using random tables and without having to locate anything in any particular place. I might put some specific things in places, but the point is that I don't have to.
I'd build another one or maybe two other nexus points that also point toward the Duke, and figure out what initial clues and information would be necessary to get the PCs pointed at least generally toward at least one of those points (like the ship and inspector examples from above). In this case one idea that suggests itself immediately is to build a nexus around drug distribution in fancy clubs to dilettante aristos, a nexus I can connect both to the farm from above, to the duke, and possibly to other nexus points.
Finally, I'll look at the nexus points and see what makes sense to connect any of those together, if anything. I'll massage the idea and move some parts around until the whole web hangs together. What you need to make this model work is a group of guilty parties. It gets very hard when its some kind of lone gunman. This example took me less than 10 minutes to bang out, and it's good enough that I'd run it for my own group with some polishing with some confidence that they won't lose the thread of events.