Let me chime in with some high level points about the RR issue.
From my point of view elegant game design, especially for puzzle games, is about having a few core features which are rich and interact a lot, so that even with rather simple rules, there's a lot of things you can do as a level designer. As a player is to figure out creative ways of using and combining the things that you know well. If you add additional features that are only used a few times, you're requiring the player to learn more stuff, only to forget about it again (and then maybe not think of it anymore when it's finally needed again, or conversely try to apply it everywhere without needing it). It can often feel tacked on. Of course, that's just a rule of thumb.
If you push this too far, you get something like all the gimmicks that were at some point in NeoLemmix, and if each level you play is about something new it breaks coherence, and in the worst case can make the player frustrated because by the time they encounter a feature again they might have forgotten about it already.
My guess is that the RR was initially added to lemmings as a substitute for fast-forward (maybe due to performance reasons), so once you're done you crank up the RR. Now one sneaky level designer in ONML figured out that it can be used for something else, and then of course custom level designers used it to some extent. Custom level designers (me included) also used all kind of glitches and particular behaviors in their levels, back in the day at least. These "features" also got removed in Lix and NeoLemmix, and their removal broke many levels. (I wouldn't be surprised if some of them were maybe more widely used than RR?)
Anyway, back in the day when fixed RR was not an option yet, I remember cheesing a lot of custom levels exploiting RR when the level designer didn't intend (of course challenge solutions also fall in that category). As a level designer it can be a nightmare to backroute-check your levels because even with a few skills, variable RR allows the player to do a lot of things which as a level designer you have to think of, and vice-versa, can provide a lot of options for the player to think about even when the level designer didn't intend for it to be used. For example, I wouldn't want any level to require RR manipulations like in that
Tame 13 solution. The long-term effect of RR changes can sometimes be hard to predict without trying out; I remember some challenge solutions required a lot of trial and error until you got everyone spaced perfectly, Tame 13 is pretty tame by those standards. Even if the level doesn't require it, as a player in hard levels you'll still think of all these possibilities, and if the level designer is unlucky you actually manage to cheese the level with that. Of course, nowadays from a level designer's perspective you can at least fix the RR, but the player still has to deal with it whenever you don't.
When the variable RR was eventually culled in Lix, not that many levels were affected in the end, like maybe 10-20 out of 500? And that doesn't mean that these levels had to be culled, many could be adapted. I'm not sure why the discussion is all focused on runners here in that regard, to answer Simon's question:
Hmm. I feel there must be better game designs than either VSI or runner to bunch the rodents.
How about a good old digger's pit, for example? Why do you want to cluster your rodents in the first place? Unless you're trying to get everyone through a fast triggered trap, that's probably good enough (and at least Lix has slow traps where this would be sufficient too). If you really need perfect clustering, how about squeezing between blockers, or as a level designer set the RR in such a way that it happens automatically? If you don't want it to be too obvious, maybe make the player dig the pit to cluster the crowd themselves (digger tunnels have a fixed width), or place a cuber to get the right width for the pit? Clustering as a level solution tool is not specific to variable RR. In the end, if you look at the levels where the use of RR can not be
reasonably elegantly replaced, you're probably down to the (low?) single digits now. It doesn't live up to even the least useful skills, in terms of required usage. And then the question is, is it really worth it to require the player to learn a new feature (and one that's not that straight-forward, to boot) just for that? If you could choose one single feature to add to Lix, would it be this? Not cloners, not hang gliders, not whatever other more useful features there could be that would qualify as a core feature in the sense I introduced at the beginning?