Yes, have the /PAE switch on both nodes. AWE enabled is enabled too.
I have set a whole lot of performance counters, don't see anything abnormal there.
"User Connections" are extremely high, more than 300 at the time the server hangs.
Yes, users first hit IIS which talks to SQL Server. When you say check the app server for "pool memory issues" are you saying use the 'Memory' performance counters on those servers to monitor?
Here is part output of dbcc memorystatus (Converted to MB/GB):
Stolen 20MB
Free 173MB
Procedures 663MB
Inram 0
Dirty 789 MB
Kept 0
I/O 0
Latched 138
Other 18.17GB
Commited 19.78GB
Target 19.78GB
Hashed 18.94GB
Thanks.