I don't mean to lecture, but this strikes me as an instance of premature optimization. What
sort of perf hit are we talking? 1%? Less? When you say that a feature will create a slight perf hit, most people probably imagine something like 3 to 7 % slower. If it's in the 1% to less range, I'd call it a negligable perf hit.
Right now you're probably doing a point vs. axis aligned bounding box collision test, right? When you probably want to be doing a circle vs. axis aligned bounding box collision test. How many extra calculations are required? I think it's on the order of maybe double for a naive test. Maybe less with a sophisticated algorithm. Maybe the
SAT.
Of course, it all takes time to implement, so if implementation time is the deciding factor instead of perf hits, that's another matter entirely.