Aims. The formation of massive stars remains poorly understood and little is known about their birth multiplicity properties. Here, we aim to quantitatively investigate the strikingly low radial-velocity dispersion measured for a sample of 11 massive pre- and near main-sequence stars (sigma(1D) = 5.6 +/- 0.2 km s(-1)) in the very young massive star forming region M 17, in order to obtain first constraints on the multiplicity properties of young massive stellar objects.

Methods. We compute the radial-velocity dispersion of synthetic populations of massive stars for various multiplicity properties and we compare the obtained sigma(1D) distributions to the observed value. We specifically investigate two scenarios: a low binary fraction and a dearth of short-period binary systems.

Results. Simulated populations with low binary fractions (f(bin) = 0.12(-0.09)(+0.16)) or with truncated period distributions (P-cutoff > 9 months) are able to reproduce the low sigma(1D) observed within their 68%-confidence intervals. Furthermore, parent populations with f(bin) > 0.42 or P-cutoff < 47 d can be rejected at the 5%-significance level. Both constraints are in stark contrast with the high binary fraction and plethora of short-period systems in few Myr-old, well characterized OB-type populations. To explain the difference in the context of the first scenario would require a variation of the outcome of the massive star formation process. In the context of the second scenario, compact binaries must form later on, and the cut-off period may be related to physical length-scales representative of the bloated pre-main-sequence stellar radii or of their accretion disks.

Conclusions. If the obtained constraints for the M 17's massive-star population are representative of the multiplicity properties of massive young stellar objects, our results may provide support to a massive star formation process in which binaries are initially formed at larger separations, then harden or migrate to produce the typical (untruncated) power-law period distribution observed in few Myr-old OB binaries.