I know how PoS was "supposed" to work
There are serious reasons to doubt this. 
but basically it's crap now without some mayor changes
It works as it was designed, and there is no crap except constant values which were chosen by hand of original developer. PoW blocks target spacing must be higher or at least equal to PoS blocks target spacing, because this system is originally designed for "slow" PoW blocks and "fast" PoS blocks. It doesn't work in the perverted version and there is nothing strange here. Orphans is just another part of self-regulation in this complex system. If developer tries to set PoW spacing lower than PoS spacing then network just forces him to get acceptable PoS/PoW ratio.If you wish to have a "fast" PoW blocks then you need another system, because this system won't allow you so. You need cunicula's concept or something like that. I.e. no PoW and PoS blocks, but a homogenious chain with a new block type. It's even possible to reuse the current code by removing PoW blocks support and forcing PoS blocks to have a suitable header hash.
PoS is allowed with wallets that have just 2YAC balance
Such threshold looks like allowing the PoW generation only for miners with 8 or 10+ MH/s. I mean it's pretty senseless and won't change anything. This measure is only able to increase an extent of the problem by decreasing active inputs amount.Maybe PoS minimum stake should depend on PoS difficulty, so small amounts are never staked if there's no need to.
Recursion here, you can't define an argument of difficulty function as a function of difficulty. 