I routinely use some PA-RISC machines (I've got two C3600 and a J6750) and while they are not "critical" systems I treat them as production systems. I run a lot of testing and internal services on them for my business. They're rock-solid hardware as you noted and run all the recent linux stuff (I'm using Gentoo Linux with linux-vserver to get virtualized servers, tracking latest kernel and currently using 2.6.34). So you're gonna get some satisfaction from them if you don't ask for fast CPUs (but they've plenty of I/O on pci buses and space for lots of ram).
The ext3 filesystem is byte order and word width independent, you will be able to mount it on every x86 x86_64 ppc or whatever else system you can run linux on, so you can sleep without worrying about a fried motherboard.
- you can find cheap HPPA hardware out there (ebay and so on) for spares if you ever want them
- the onboard SCSI controller doesn't exactly scream "high performance" in todays terms, and SCSI2 disks are slow, small, and expansive. You can use internal disks for the root filesystem (a pair of disks in RAID1 software will protect your uptime from a disk failure) and run workloads off external storage (USB2, firewire, gigabit eth...) or even go 100% diskless with a network boot. A 64bit / 66MHz Intel gigabit ethernet will get you good performance combined with the bus capacity of those systems.
For the record, I run my systems with root and tmp and swap on internal disks (10kRPM 9GB or 18GB disks in two disk RAID1 on each system) and use USB2 external disks or gigabit ethernet for the actual work data. Uptimes are in the two-three years range if you don't count reboots for new kernels. I'd like to transition to diskless when I'll have time and get rid of SCSI disks alltogether, using storage space from nearby SATA or SAS boxes.