Strange speed test results on different home computers

Unsure if this is the best sub-forum but I think it relates more to my home network than the Internet connection...

I have a home server that does periodic speed tests using cURL. What I am seeing (consistently) is that some test providers achieve a substantially lower speed than others, but on another machine on my network I can reach the full speed of my Internet connection.

For example, Think Broadband (IPv4+6) is 7Mb/s on the server (FreeBSD) but 51Mb/s on my Mac desktop. However OVH (IPv6) and Vodafone (IPv4) can reach 53-58Mb/s on the server. Another example that is slow on the server is OVH (IPv4) at 12Mb vs 52Mb locally.

Both machines are connected via Gigabit Ethernet to a switch which then connects to my ISP router (Sky).

Here is a graph showing the speed tests:
smithers_speedtest_20220702.png

The jump on 29-June is because Openreach came to fix the connection and the sync speed went up from 38Mb/s to 65Mb/s.

For reference, I have the same test on another server on a different Sky connection and the results look fine:
burns_speedtest_20220702.png

Any ideas what could be causing the lower speed on my home server and how to go about fixing it?

I didn't get to the bottom of this but I moved to Hyperoptic this week and the graph now looks like this:
smithers_speedtest_20220709.png

So still seeing discrepancies, e.g. OVH is 259Mb/s on the sever but 592Mb/s locally and Hetzner is 252Mb/s vs 546Mb/s but the server is clearly capable of more, e.g. 702Mb/s on Think Broadband.

You may no longer be testing your connection but the limits set on the server for the path to you. Eg faster speeds to their own network.

I think I have solved this.

When testing on HTTPS I can see cURL is pinning the server CPU at 100%.

It's Vodafone and Think Broadband whose test files are served over HTTP that have faster speeds.

In reply to a post by Cockroach:

When testing on HTTPS I can see cURL is pinning the server CPU at 100%.

Entry level and slow CPUs can have more of an impact from encryption, often found as shared CPUs in VPS platforms.

It's a 10 year old HP ProLiant MicroServer, so not a great CPU.

In reply to a post by Cockroach:

It's a 10 year old HP ProLiant MicroServer, so not a great CPU.

Ahh, you'd probably do better with a CPU that had AES instruction acceleration, e.g. Intels AES-NI that came in with some of the 2nd Gen Core i5/i7 desktop CPUs.