This can depend on several factors:
It is possible for the device to maintain the duty cycle, or transmit some downlink data at the same time, or not demodulate the data due to interference.
There are many more reasons.
However, this does not necessarily mean that messages are lost because packets can indeed be received by 2 gateways.
Duty cycle limitation is applied to transmission, not reception
Yes. An 8-channel gateway is actually rather complex, it can receive on 8 different frequencies, it can theoretically actually receive two signals at different spreading factors on the same frequency at the same time.
At some stages of the receive process there are a number of dynamically assigned “engines” which can work the problem, and if they are all assigned chasing other possibilities (even false ones), a possibly valid signal might not be investigated.
Having a node extremely close to a gateway can actually increase the chances of this, because strong signals can weakly bleed over and appear as weak (possibly unintelligible) false signals on other channels. This could increase the number of distractions the gateway chip’s processing engines have to contend with in finding the valid signals.
Duty cycle limits theoretically apply to both types of transmitters, yes. But they only apply to transmitters hence they cannot explain why two gateways report a different subset of packets received from the same node
(Incidentally while duty cycle is thus irrelevant to the question at hand it’s also worth noting that in some regulatory schemes, 500 KHz downlinks as a gateway may transmit are considered themselves spread spectrum and so may have looser duty cycle rules, vs. the more narrowband 125 KHz uplinks sent by nodes. It’s not that either device couldn’t send the other kind of signal, but as the existent gateways can only receive multitiple narroband channels (and one wideband one) typically nodes transmit narrowband and gateways transmit wideband, but practice and regulations vary by region)