Think we found the problem. Host1 was connected through SAS on Storage Processer A and Host2 connected to Storage Processor B. It seems it works like with EMC storage and the LUNs kept thresspassing. Once we connected host2 to Storage Processer A the performance was back and there were no more messages in the vmkernel log.
Stupid we didn't think of this before because I usually do check this with EMC storage. Just not used to IBM SAS Storage.
Thank you for your help.