So even under the ridiculously unlikely assumption that computational power halves in cost every 18 months for the indefinite future: 50 years from now, the entire current global GDP still won't be enough to buy a processor that can crack a Bitcoin key in less than a decade.
(Of course, this is assuming there are no other unknown flaws in the protocol, and that no better algorithms are found to find discrete logarithms over elliptic curves. But if a vulnerability is found, it won't be thanks to Moore's law.)
[1] http://en.wikipedia.org/wiki/Discrete_logarithm_records#Elli...
As for ECDSA, it can only be broken by quantum computers, but still most balances are protected by sha256*ripemd160 hash, so they are safe in the face of sudden QC attack or a weakness in ECC. People will want to protect their wealth and thus will change the protocol for other algorithms.
But do you think it would actually be viable to make a protocol algorithm transition?