If they were forking that JVM in order to offer it as a development tool to a broad developer base, then such an optimization might be worth keeping.

Someone out there might have an application in which they are serializing large numbers of large integers.

The case for abandoning it is not as strong, since you don't know who is doing what.

It's a guessing game in programming language run-times and compilers: "will anyone need this improvement?" You have room to invent "yes" vibes. :)

If the distribution of numbers you are serializing is uniformly random, you generally wouldn't choose a variable length encoding. Sometimes the choice is made for you of course, but not usually.

The problem is that this is a complex assembly implementation that needs to be maintained over the course of decades in the future. Unless you have a good reason to keep it, you should drop it.