Yes, we are monitoring everything in LWIP_STATS as well, and while we do see peaks in other resources, the only one we seem to be losing is PBUFs.
Yes, it does appear to be pool pbufs we are losing as we do not appear to be losing heap, and we are monitoring lwip_stats.pbuf.used -- which is keeping track of the pool pbufs.
Are these only used for reception? I'm not very familiar with how they are used in the stack, although I was just about to go investigate that ...