The first complaint to me seems invalid. Lets assume we have a data center with 900 servers split into three groups, or sub-domains. Say a NIC on a server in group 1 starts going bad and flaps causing a flood. It doesn't matter what kind of design you used for the network in the data center, every device in the broadcast domain will be affected by that flood unless you have some code running on networking devices that look for such things.
If you have a tree diagram or something else, it doesn't matter. Network design has little affect on what server needs to be in which network or VLAN. But this should be a moot point anyway as all Cisco switches sold today can be programmed to shut flapping ports.
Point two is valid. The only thing I can really say is that if the network admins budget correctly they can minimize the effect of this. The company I work for has a 4 year life cycle on all networking equipment unless Cisco "End of Life"'s it before the 4 year period. This allows us to keep the network up to date without spending a lot of money. Your situation may be different.
I think the advantages of the tree far outweigh any negative one can come up with. The redundancy and speed (if you collapse layer 3) can really pay off dividends in a data center type environment. The only way to get more speed and redundancy is to go with a mesh environment, but that can get near impossible to troubleshoot and is very very expensive and rarely used in a LAN design. Sometimes mesh is used for WANs, but still rare due to troubleshooting and cost.
I hope I helped, let me now if you have further questions...
