Recently I noticed that cura always is giving me less printing time than the print itself takes on my tevo tarantula with Marvin firmware.
The time difference is about 15%, at requested 50mm/s printing speed.
How could I verify real printing speed?
Printing speed is dependent by the firmware and physical properties of your printer.
Slicers typically compute the expected time by assuming the printer will execute exactly what it is instructed to do, but a printer is a real object, with mass and momentum, and stepper motors that have an upper limit for their power output and rotation speed.
So for example, the GCODE may say "extrude 200mm at 100mm/s" and the slicer will compute that operation as taking 2 seconds. However the printer will need to accelerate and decelerate at the extremes of the movement, and it may even be incapable of reaching speeds over 70mm/s, so the actual operation will likely take 3 seconds or more.
Accelerations and decelerations account for most of the difference between ideal time and real one, and since the number and intensity of those is totally dependent by the GCODE/model being printed, it is not possible to simply multiply the computed time for a given factor (for example
1.15, as your question seems to imply). A large cylinder printed in vase mode will have a printing time much more similar to the computed one that an intricate model with a very complex surface, for example.
In recent years, slicers that are maintained by a printer manufacturer (cura, slic3r PE) have become better at estimating printing times for their own printers, as the settings of the firmware are accounted for in the actual estimating algorithm.
If you use Octoprint, you may have noticed that the time estimate octoprint gives improves over time, as octoprint will analyse the GCODE and measure the elapsed time, and will be able to guesstimate the real time with an increasingly degree of accuracy.