We use IBM networking (used to be BLADE networks) Rackswitch 8264. They will do TRILL, and have multi-chassis link aggregation, they call vLAG. We use this for cross datacenter aggregation. They do have the L3 features you are looking for and BGP as a possibility, but no full tables. It is a cut-through switch (although this can be toggled in software to store and forward in later switch os). I believe, although I can't find the doc where I read this at the moment, the packet buffer is 2G, but shared among ports.
Enterasys S-Series is also an option, but the 10G port densities are much lower. S-Series has large packet buffers, chassis bonding, and L3 features (some modules support full bgp tables).