The situation may differ greatly among fields? For CS, ACM digital library is $99/yr which is pretty reasonable (I've been paying it for years and willing to chip in that range even ACM allows the author to put up the submitted version on his/her own site for free). I do occasionally hit paywalls of other publishes, and wish they make non-member per-article fee cheaper---like $5 or $10 instead of $35; they might set the per-article price that high to encourage institutions to sign up the package?

I'm not a researcher though, so my pursuit of papers are mostly purely interest-driven and can be given up easily. If you need to survey large number of papers I'd imagine the fee quickly accumulates.