I also wonder if it's just harder to continually monitor performance in a way that alerts a team early enough to deal with regressions?
That said, security _can_ impact performance! I work on Stripe frontend surfaces, and one performance bottleneck we have comes from needing to use iframes to sandbox and isolate code for security. Having to load code in iframes adds an additional network load before we can get to page render.
Over time decisions are made independently by devs/teams which cause the code bases to get out of alignment with a performant design. This is exacerbated by the growth pressure. It's then really hard for someone to come in and optimize top to bottom because there is everything from a bug to a design problem. Remediation has significant overhead, so only things around the edges are touched.
Fast forward a couple of years and you have a code base that devs struggle to evolve and add features to as well as keep performant. The causes are many and we come to the same one whether we complain about performance or maintainability. You probably don't feel this way, but Stripe and other premier engineering companies are way ahead of others in terms of their practices when you compare with the average situation developers are facing.
Independent mobile development is often where I see most attention to performance these days. The situation for these devs is a little bit closer to what existed in the nineties. They have a bigger span of control and performance is something they feel directly so are incentivized to ensure it's great.
I remember everyone complaining about how bloated they were at the time. Pretty sure someone in 2055 is going to run today's Office on 2055 computers and marvel at how streamlined it is.
https://copy.sh/v86/?profile=windows95
Once it boots, run povray, then click run.
It took over 2 minutes to render biscuit.pov, though it did manage to use SSE!
We used to wait two hours for a mandelbrot to display on a Commodore 64, and were delighted when it did.
Ironically, it was specifically the longer feedback cycle of long builds and scheduled releases that seems to specifically have given us better software?
You can run Office 97 now and it'll start fast because disk i/o and cpus are so much faster now. Otoh Excel 97 has a maximum of 256 columns and 64k rows. You might want to try Excel 2007 for 16k columns and 1M rows, and/or Excel 2016 for 64-bit memory.
You could build fast software today by simply adopting a reference platform, say. A 10 year old 4core system. then measuring performance there. If it lags then do whatever work needs to be done to speed it up.
Personally I think we should all adopt the raspberry pi zero as a reference platform.
Edit: * office 2000 was fast too with 32 megs of ram. Seriously what have we done?
I rebooted a lot though (mostly used Linux by then), so maybe it was just fighting with itself when I immediately fired up office after boot.
It seems fascinating how much more efficient Windows apps were back in the nineties, capable do to almost everything the same today apps do in a similar manner on orders of magnitude less powerful hardware, often performing even faster.
The last time I expressed this, probably also here, somebody suggested the performance drop is the cost of modern security - vulnerability mitigations, cryptography etc.