Some newer hardware features are introduced to significantly improve broadcasted TLB invalidation performance notably AMD INVLPGB [1] and (eventually) Intel RAR. However, as highlighted in [2] this is currently tricky to implement safely in Xen due to pagetable handling limitations. The idea would be to first identify those blockers and find a approach towards fixing this limitation (at least for Xen own use)
[1] https://www.phoronix.com/news/AMD-INVLPGB-Linux-Benefits
[2] https://github.com/xen-project/xen/commit/5b718d24e88ceb2c28010c647836929b85b22b5d