I'm not sure about books, because the 'Net tracks everything so much more effectively. Treating the internet like a firehose-book, then...
- If you're interested in Linux kernel activity in particular, sign up for LKML updates so you have a constant feed of "ooh, what's that" to keep up with, although following everything in realtime might be a bit of an overload.
- It's good to have a passing idea of how the major distros work - if you have enough familiarity with a system to at least comfortably spin it up in a VM easily and quickly, it makes it that much easier to test how the occasional obscure edge case is handled in a given environment. If you're writing at the systems level (eg, background daemons, systemd vs sysvinit et. al.) this will probably come up reasonably frequently.
- I've learned that some languages can actually be almost as fast as C nowadays - in http://www.gopherjs.org/blog/2015/09/28/surprises-in-gopherj..., a small test algorithm that computes pi using 1 billion iterations runs in 6.434s when compiled with gcc -O3, and the same code written runs inside Node.js in 6.549s - 105ms difference, arguably nil.
- Real-world applications do undeniably introduce latency in the most optimized of environments, but capable, speedy CPUs are approaching such ubiquity that scripting languages are an extremely viable choice for a lot of tasks.
- I understand that IllumOS - the now community-maintained open-source project that used to be Solaris - has awesome kernel-level debugging facilities. This might be interesting.