Ok, I replaced the bearings with self aligning ones and got the motor running.
Everything seems fine now and I can run the motor for as long as I want with no change in speed. So I went on to measure the efficiency.
After gathering some data and calculations, the numbers seemed bad. I loaded the motor 8mm shaft several times and each time with different force, but all the calculations showed efficiencies 31% and below
For example, here is data from one of the tests:

Input power: 25.5V, 2.47A.
RPM's 4146
Scale deflection: 1.28Lbs on one scales and 0.71Lbs on other scales.

So the input power is 25.5v x 2.47a = 62.99W or 0.0844 Hp

The shaft diameter is 8mm so the circumference is 8xphi or 8x3.1416=25.13mm or 0.0824 Ft

RPM x Ft / 60 = 4146x0.0824/60=5.69 Ft/s

5.69Ft/s x (1.28Lbs+0.71Lbs)= 11.32 Ft-Lbs/s or 0.0205 Hp

So Efficiency is output/input or 0.0205/0.0844=0.242 or 24% efficiency.
Is this right? Seems too bad. The other results are similar.

Maybe I measured the deflection on scales wrong? I did this the following way. I put the motor so that the shaft end is reaching past the table top, then I attached a piece of thin cloth sheet to the scales. I turned the scales ON and just let them hang on the shaft using the cloth sheet. To the load was just the weight of the scales. This way I could get steadier results, because it is difficult to load the motor evenly if you hold the scales above the motor in your hands. Maybe this is the wrong way to do this?
I am a bit confused about these results