Slight performance regressions in recent PRs

[smarter]

Looking at http://dotty-bench.epfl.ch/ it looks like we've become about ~2% slower on some of the tests recently. It's hard to see due the variance but I think the following PRs are responsible for this:

• #3634: A symmetric meta programming framework by @odersky
• #3575: Reduce the depth of trees generated in PatternMatcher by @nicolasstucki

I'm opening this issue because I think it would be nice to know why these PRs had a performance impact and whether this can be mitigated in some way. 2% isn't a big deal so if we can't figure it out we can just close the issue.

[odersky]

This will be fixed by #3866.