Different indoor speed with the same setup

the Zwift speed is based on power/draft/route ratio, you can't adjust a wheel size accurately because the draft is depending on the number of the riders around you during the workout, it is totally random thing (the route could be absoluty empty, or with light, medium, heavy and absolute "heavy" traffic if you go all the time with a robopacer and with 50-200riders, lets say with 200watts i can ride with 36km/h avg on Tempus if i am absolutely alone, and 43-44km/h also if i go with the maximum draft. and the wheel speed on my bike has much less difference or nothing ) so even on Tempus Fugit you cant set the wheel size right... so too many variables what you can't adjust , the 2 speed values are computing from 2 totally different sources (and with different variables), so there is no right match for ALL circumstances (only if you ride always totally alone and on the same route, and with 100% trainer difficulty because it has effect for the gear ratios which effect for the physical speed too)

DO NOT MEASURE (or at least ignore it...)  the indoor trainer's speed if you ride on Zwift (or in any virtual app) , what is matter : duration and power, nothing else.

you ride on Zwift, not? and record your workout with the Edge 1040 too and your trainer is paired to the Edge .

25 vs 35km/h. Was the used gear ratio is the same? (small ring front vs big ring front, and what was on the rear)  Was it free ride or ERG mode training? 

The both fit files show the same origin of speed data: your Tacx trainer.

What exactly do you mean by "old speed"?

StefanScheddy said:

but sometimes it's dropping back to the old speed

yes, it is Neo2T. i am not sure that you can modify the wheel size or if you modify it, is it stored or not (especially if you unplug the Neo) Honestly i never tried to modify this value since 2015 :D, because an unnecessary and not real data. 

What is your Neo ANT ID? 27907, right? both file has 3500mm as set. so i assume you were on different gearing. 

Me, too, I am riding a Tacx NEO (first generation) and Tacx Trainings App (PC).

Just to explain:
I'm usually riding films and record it with PC app and in addition with my Edge 1040 to keep track of fitness data in GC (I delete this rides afterwards and keep only fit files from the PC app).
I've connected my NEO as Speed sensor to my Edge and set the wheel circumference to 3000 mm, as this value meets the speeddisplay of my Tacx PC app at a speed around 20km/h the best.
( End of explaination)

What I found out:
-For plain rides the speed (and therefore distance) meets the expectations.
-For climb rides, when speed comes down to 5 km/h, my Edge shows around 8 km/h and therefore much more distance as the length of the course is in reality.

But if you ignore the data from the PC app, which we will do here, the duration, distance and average speed always match.
And they do in both of your fit files.

What I want to say is:
Climb or not climb influences the speed shown on Edge compaired to reality.

yeahm the Tacx app is a bit better because there you ride alone and its physics is based on the trainer power curve (because both products is by Tacx) but as You wrote, there are difference even in this case too . And the Zwift is a different thing, the default trainer difficulty is 50%, so with that the ingame "virtual gearing and speed" heavily different then the physical gearing and disc speed (you can climb even on big ring on Alpe because it halved the grade, i dont think that you do it in reality on the 10% :D but you can on the 5% ) 

but if the OP will answer my previous questions, probably we can guess out what going on here :)

Both rides are in ERG mode with 160W constant. I'm not changing gear, it's always the same setup. This is why I'm confused all settings are the same (even route on Zwift the same). Most times the speed is around 35km/h, but sometimes it's changing back to the old 25km/h.
Old speed is for me the setting at the beginning which was horrible low.