I'm not expert of language compiler. but I fuzzily remember Reini Urban's criticism about perl6 that because of rakudo's wrong 3-stage design, perl6 would be never as fast as perl5! Is that right?

That sounds like a very dubious claim to me. Maybe the staged approach limits the performance of compilation time, but the problems you've mentioned (working with many and large files) sounds like they are limited by run time performance, not by compilation time.

if he's conclusion was wrong, when can I use a decent fast perl6?

Such things are very hard to predict in a volunteer-driven project. You could also ask when you can expect a Perl 5 that supports easy to use and robust concurrency, or a Perl 5 with less gotchas about mixing string and binary data, and probably not get a very good answer.

There are use cases for which Perl 6 is fast enough now, and there are small niches where Rakudo on MoarVM can be faster than Perl 5 right now (for example integer arithmetic when you use native types in Perl 6).

does perl6 deserve to invest for career? or just treat it as hobby? Please enlighten me. TIA.

If you want to make money now, Perl 6 isn't your best choice. Treat it as a hobby, and learn one of the trending languages like Javascript or Python for making money.