Recent progress in the development of superconducting nanowire single-photon detectors (SNSPD) has delivered excellent performance, and their increased adoption has had a great impact on a range of applications. One of the key characteristic of SNSPDs is their detection rate, which is typically higher than other types of free-running single-photon detectors. The maximum achievable rate is limited by the detector recovery time after a detection, which itself is linked to the superconducting material properties and to the geometry of the meandered SNSPD. Arrays of detectors biased individually can be used to solve this issue, but this approach significantly increases both the thermal load in the cryostat and the need for time processing of the many signals, and this scales unfavorably with a large number of detectors. One potential scalable approach to increase the detection rate of individual detectors further is based on parallelizing smaller meander sections. In this way, a single detection temporarily disables only one subsection of the whole active area, thereby leaving the overall detection efficiency mostly unaffected. In practice however, cross-talk between parallel nanowires typically leads to latching, which prevents high detection rates. Here we show how this problem can be avoided through a careful design of the whole SNSPD structure. Using the same electronic readout as with conventional SNSPDs and a single coaxial line, we demonstrate detection rates over 200 MHz without any latching, and a fibre-coupled SDE as high as 77%, and more than 50% average SDE per photon at 50 MHz detection rate under continuous wave illumination.