No, it's not, that's the point.
Md5 has 128 bits but only 40 bits of space that can realistically be used for such matters.
SHA1 doesn't seem to have had its keyspace reduced in any meaningful way. It hasn't suddenly become less secure overnight. If it takes 6,610 years of computing time to produce one collision, where exactly is the problem?
The real risk is, and always has been, rainbow tables. Which this latest development does nothing to address, because it doesn't change the effectiveness of rainbow tables which for forums are the real problem. This change doesn't suddenly make it easier to make rainbow tables, nor does it make it quicker to find collisions in them.
And since SMF has always salted the hashes at a per user level, the argument is even more disingenuous for arguing the value of hashing.
Would I ever use md5? Back in the day before people realised how insecure it was, everyone used variants of md5. We now know better.
SHA-1 does not "suck as much as md5" for passwords, mathematically provably so. md5 can collide as frequently as every 2^40 hashes (approx every 1,099,000,000,000 hashes) owing to its weaknesses while SHA-1 still collides every 2^160 hashes (1.46E+48 hashes, so, 1 with 48 zeroes after it) as we haven't seen any proof yet that the keyspace has been reduced, but don't let facts get in the way.
By definition there are going to be collisions in hash functions because there is an infinite number of possible inputs and only a fixed size of output so you are guaranteed to get collisions. Just as you are in any hashing function. By that logic, bcrypt is also insecure. Oh snap! Or, of course, you can realise that just because there is a proof of concept proving that's it's easier to collide an SHA-1 than previously thought, that doesn't automatically mean that it applies in every case or that every possible use of SHA-1 is somehow contaminated.
On to your other question... Should SMF move? SMF 2.1 already has. Would it not be a better use of everyone's time and energy to get 2.1 out the door rather than backporting things out of it? Then everyone wins.