It would be interesting to see HN-ers that use facebook auth in their apps share some stats on how big percentage use it? I too believe you should have your own auth system as a base, but maybe someone can provide some numbers proving that it actually is a waste.
I may very well do away with Twitter and email registration. Some of those 5% of new users might use Connect, and some might leave. That's fine. What I care about is streamlining the experience for the vast majority of my user base.
"I too believe you should have your own auth system as a base, but maybe someone can provide some numbers proving that it actually is a waste."
This is the way to look at it. Each service should test, analyze their numbers, and make the decision that makes sense for them. Blanket statements like "startups should never only use X for authentication" are just wrong.