I have encountered strange problem. I got few MikroTiks in my LAN. The main router connected to the Internet is hEX (10.10.10.1). It has to bridges configured because of the CAPs reasons. Anyway, there are two other MikroTiks connected to it: RB2011UAS-2HnD (10.10.10.3) and RB951G-2Hnd (10.10.10.4).

Now, when I run broadband test between 10.10.10.3 and 10.10.10.4 it shows almost 1000 Mbps/1000 Mbps, but when I run the same test between 10.10.10.1 and one of these two other devices (10.10.10.3 or 10.10.10.4) it doesn't show more than 500 Mbps/500 Mbps. What might be the problem?

I'm pretty sure you use btest for testing, and that is the problem.
Please search the forum, here's plenty of posts explaining why btest should not be used, and what is the proper way to test your bandwidth.

How should I search the forum? When I look for "bandwidth test" or even "alternative bandwidth test/btest" it gives back tones of posts, about btest...

Using btest on the same devices you want to test stress the cpu because it has to generate all that traffic in such way it is not able to handle to much traffic.
It's like trying to talk while you drink. It's hard.
You should use btest through the devices you want to test, that means you have to have two more devices.

Yeah? So let me explain...

1st device
hEX (10.10.10.1) - router + CAPsMAN
RAM: 256 MB
CPU: mmips 880 MHz

2nd device
RB2011UAS-2Hnd (10.10.10.3) - switch + CAP
RAM: 128 MB
CPU: mipsbe 600 MHz

3rd device
RB951G-2HnD
RAM: 128 MB
CPU: mipsbe 600 MHz

Now, when running test on hEX (the most powerful of the above three) I have 500/500 Mbps when testing to 10.10.10.3 and 10.10.10.4. When running the same test from 10.10.10.3 or 10.10.10.4 with 10.10.10.1 it shows the same (500/500 Mbps), but when testing bandwidth between 10.10.10.3 to 10.10.10.4 (no matter if I am logged in to 10.10.10.3 or 10.10.10.4) it shows 1/1 Gbps. So please, don't tell me about traffic or whatever. The CPU usage of hEX during the test is less than 50%, while on the other two reaches 98% and shows 1/1 Gbps.

Honestly, I don't think this is a problem, but I have another RB951G-2HnD that I can use, but how to do it, to make the measurement done properly?

The most proper way to test is to use two computers running iperf. Or two (not one!) additional routers running btest, but please make sure btest itself does not max out the CPU (also please note that hEX r3 is a dual-core device, whereas btest is a single-threaded app, so if you run btest and see 50% CPU utilization that actually means one CPU core is maxed out).

The test traffic should always go through the router you are testing.
The test traffic should never be originated from nor terminated on the router you test.