I was having a debate with a colleague about the merits of requiring or just allowing perhaps users to receive results via e-mail. Personally I've never seen the point of receiving results by e-mail, and definitely never seen the point of a website requiring it.

The disorder predictors available offer a case study in the effectiveness of such measures.
IUPred (no restriction) 246
Pondr (restricted) and incidentally the best 94
Disopred (restricted webserver but allows access to DB of preprocessed results) 157
RONN (no restriction) 188
Globplot (no restriction) 380
Disembl (no restriction) 396

I think this shows pretty nicely makes the point that requiring an e-mail is bad for citations.