I wouldn't use "with" in most situations, but I like it in the testing code and for some DSLs, as it makes things more readable (again, personal style).

I think that once you start optimizing performance problems you need to 1) measure things anyway, and 2) fix on a case-by-case basis. (Don't optimize prematurely, etc.)

In addition I'd say that the interpreter is broken and should be fixed to be faster. "with" is not deprecated.

The trouble with armchair quarterbacking is that many things are easier said than done. It's not like they're deliberately making your code slower — the with-statement makes certain optimizations unfeasible. Their options are to always disable those optimizations or to do it only when there's a with-statement in scope — there's no third option that makes with-statements magically fast. If you can come up with new optimizations that make the with-statement faster, I doubt they'd turn down your patch.

I don't believe that this is just an oversight because "people say you shouldn't use `with`". It simply cannot be optimized in the way that normal property accessors are. If you look at old IE perf, they're actually pretty close (both really slow).