The variable configuration of Raman spectroscopic platforms is one of the major obstacles in establishing Raman spectroscopy as a valuable physicochemical method within real-world scenarios such as clinical diagnostics. For such real world applications like diagnostic classification, the models should ideally be usable to predict data from different setups. Whether it is done by training a rugged model with data from many setups or by a primary-replica strategy where models are developed on a 'primary' setup and the test data are generated on 'replicate' setups, this is only possible if the Raman spectra from different setups are consistent, reproducible, and comparable. However, Raman spectra can be highly sensitive to the measurement conditions, and they change from setup to setup even if the same samples are measured. Although increasingly recognized as an issue, the dependence of the Raman spectra on the instrumental configuration is far from being fully understood and great effort is needed to address the resulting spectral variations and to correct for them. To make the severity of the situation clear, we present a round robin experiment investigating the comparability of 35 Raman spectroscopic devices with different configurations in 15 institutes within seven European countries from the COST (European Cooperation in Science and Technology) action Raman4clinics. The experiment was developed in a fashion that allows various instrumental configurations ranging from highly confocal setups to fibre-optic based systems with different excitation wavelengths. We illustrate the spectral variations caused by the instrumental configurations from the perspectives of peak shifts, intensity variations, peak widths, and noise levels. We conclude this contribution with recommendations that may help to improve the inter-laboratory studies.