Perhaps the major issue with OSPF is its usage of the Dijkstra algorithm. This algorithm doesn't scale well, which is why there's often so much emphasis on area design with OSPF. Often not fully understood is how much network topology can be safely contained within an area, which can be more involved then just having X number of OSPF routers within an area. When these factors are ignored, I have seen OSPF meltdown from a link flap.
Another issue with OSPF is its costing capabilities. If you have a large spread between the fastest links and slowest links, OSPF might be unable to indicate different path costs if the accumulated path cost "overflows", so for instance a T1 link might be costed the same as a fractional T1. This is not the issue of what Cisco's OSPF is using for a base cost metric, although it can be related to it. (On that subject, believe Cisco sets a base cost of 1 for 100 Mbps, so without adjustment or override, 100 Mbps and faster links all appear the same cost.)
[edit]
PS:
My understanding of the OSPF standard, some issues are not fully defined and left to the implementation. (Sort of like how big an INT is in C.) My experience has been Cisco's implemention has been modified over the years to better handle some corner cases. Other vendors implementations are not always as robust. In other words, I've seen a condition crash brand X but not crash Cisco.
