I think there's two variables to consider.
the first is any serial operation that done on a LUN level. in modern OS (like Cdot) the serial operations will be to much smaller chunks, but in ONTAP pre 8.1 and in other vendors, some serial operations are bound to a LUN. which could create a bottleneck.
The other variable is the SCSI queue depth. in VMWare there's a 32 deep queue per path - per LUN. if you see the queue getting bigger and bigger, you should either add path's or split the workload to multiple LUNs (assuming there's no underling bottleneck that causing the queue growth)
there is also a theory that you better move workload with sequential workload away from random ones, as it will perform better for the sequential and prevent the random IOs on the same LUN to wait behind for the sequential.
Gidi