50% was at the time when you were lucky to have more than 2GB RAM...
Plenty buffers is good- too much can be a pain... the best results is when dbc_max_pct does not exceed 2.5GB so in the present case that would be something like 8%
reducing dbc_min_pct makes sence when you have memory shortage - Not the case here but could be reduced to 4%..
The issue I see but you havent mentionned it happned yet is more :
" Why cant I load more programs (no more memory error message...) when I have far more RAM than what is actually used???"

This is more what could happen to you, why?
Because you are using pseudoswap... HP-UX ( at least 11.11 ) is a OS using Virtual memory, to make it simple it will load first in swap to "reserve" the space then load in memory, in case of need to swap the space already reserved is used so in practical you could have far more program running than actual RAM only system will be slow because using its swap...
Your case is your swap device is far less than actual RAM, with pswedo-swap was not activated you could actually only use at maximum memory to the size of your swap: 16GB when you have 24! so turning swapmem_on to 1 lets you use the excedent RAM as swap so you can use more than 16GB
But to be able in case of crisis ot use all the RAM at other things than swap, you would have to create a secind swap device for performace - of the same size of the one you have: 16 GB bring your system to have 32GB of virtual memory.
Having 2 of same size will bring the OS to use both simultaneously (like load balancing) if it needs to
About SGA buffers etc thinking to make them bigger will be better can lead to performance surprises...
I have been called by a bank years (more than 10...)ago when they had performance issues when changing to batch mode just like you I asked for a quick sysdef and how much RAM they had... I knew the type of disks (FWSCSI...) so I asked them if it was something like the system freezes so often say every1/4 1/2 hour for a couple of minutes...
They wanted to know how I knew... (a full buffer of 5 GB will take more than 4 minutes to sync with those disks - the bigger the buffer , the more time - and its not linear but more like exponential hehe)
I said we could tune the JFS but a good start was to compile a new kernel with new buffer values, then see how much it improved, it solved their issue...
As fo SGA I have seen SQL request 30% faster on systems with 50% less SGA -The same! To much buffers to scan when data is not there will slow the system when with less SGA it would already have gone reading the disks

You said you are on a SAN - The SAN normally has huge cache also, so why waste resource by using sequencially 3 huges buffers before actually going to access the disks?