Unless you are calling the subsequence millions of times in a row then chances are there is no performance difference between Current File and an actual sequence file. The performance hit you would take is if the step is set to load dynamically as opposed to load when the sequence file opens.
So the only time you want to move subsequences to a different file is if they can be reused by other sequence files. Essentially creating a Sequence Library (that's what I call it). So let's say you have ProductA and ProductB. You create a test called ProdA.seq and you create a test called ProdB.seq. Now let's say that both tests want to perform a current draw test. And because you are a good modular developer you have a subsequence that represents the current draw test and all of it's steps. You move this subsequence to a shared sequence file, we'll call it CommonTest.seq. That way both ProdB.seq and ProdA.seq can call it. Then if you change the complexity of the test or fix a bug then they both get the changes.
You could call the current draw subsequence from ProdA.seq by ProdB.seq but then what if you want to just put ProductB on a bench by itself. Now you need the code for ProdA too.
Anyhow, hope this helps,