I don't understand what you're confused about.
Look, from your prior comments you've set dbc_max_pct to 20. Assuming you left dbc_min_pct to 5, that means that the system will use between 5% to 20% of total memory for buffers in the buffer cache to improve performance. 20% if there is more than enough free memory that it would be sitting around idle, and down to 5% *when there is memory pressure*. 25% of memory free (or 50% in this output) is not memory pressure. (Closer to 1% free). Idle RAM is wasted RAM.
So -- according to the first output, your kernel is consuming around 20% for kernel dynamic data and another 20% for DBC. Since you told it it could, that's to be expected. (Obviously there's actually some buffered I/O going on to generate these). 50% or so of the memory is free, and the remaining 10% of the memory is User (7%) or Static system.
Within the 20% of the kernel dynamic memory allocations since you chose to break it down -- 4% is in the superpage cache of the kernel [free kernel dynamic memory pages which are subsets of large page kernel translations for improved Translation Lookaside Buffer (TLB) performance -- if you can access multiple kernel addresses without causing a TLB miss fault on the processor, you get better performance. These will be used for new allocations or returned to the system when entirely free (if Garbage Collection runs... which it _won't_ with this much memory laying around. Again, HP-UX favors actually using your memory for performance until you need it for something else.] Another 3% is in metadata needed for the buffer cache... the rest (11%) is split across general metadata and other stuff needed for the kernel to support what you're doing, but dominated by VxFS related metadata -- so I have to assume you're doing a bunch of read/write operations across one or more (probably large) file systems... hence you cause the kernel to allocate file system metadata, buffers for the read/writes, etc...
This all looks perfectly normal and fine. If you started putting an actual load on the box such that the User component rises and free memory goes down to 1%, memory will be reclaimed from the Buffer Cache, kernel garbage collection will run as needed, etc. to flush some of these performance enhancing caches to get you memory for your user space activities.
