I was testing ping on the LAN from main router to WiFi clients and I have got some high response: from 2ms to 110ms, with several variations.
I have tested phone, tablet, laptop all near by the AP (same room).
I have 2 setup under test (2 different sites): one using VLANs and the other on a flat LAN.
On both setup I'm using CAPsMAN, which is set to local forwarding on both cases.
Just to check that CAPsMAN was not the issue, I changed one AP manually and I have got the same end result with high latency ping.
If I connect a PC directly to ETH on the same AP, latency is as expected 0ms and ping-arp works fine; while from WiFi I receive a lot of timeout, packet loss when using ping-arp.
I don't think it's a VLAN setup issue on the AP, as the other site is plain no VLAN, but same response on their LAN and WiFi clients.
In general I don't have issues with the clients, in fact I was testing ping today for other reasons when I got caught by this result.
Honestly I have never checked ping on WiFi side to clients, as all was appearing normal based on RF signal and behavior, so I don't have any reference to judge.
But spreading latency up to 100ms it seems too high considering that I have got 10ms on WAN when I check internet destinations.
On both APs and main router I don't see CPU going high; the main router is RB4011 on one site and CCR1016 on the other. All devices are running on long-term 6.47.9.
May be is something obvious, but I'm little bit puzzled about this.
Thanks in advance for any suggestion to investigate the root cause for this.

Edit some RF info: from the WiFi, it's 5GHz network on non-overlapping channels, no interference or radar issue. Clients get good speed overall, signal is relatively strong and I can see on the 80MHz channel width that phone or tablet gets 650/780Mbps as connection speed and 2S on all cases.

Wireless clients (most notably battery-powered ones, such as phones and tablets) frequently enter sleep mode and during sleep they don't listen to radio. AP knows that and buffers unicast packets until devices wake up. This way packets get delivered, but delay has high jitter. So if you want to test wireless RTT, you should run ping from such wireless device (this way radio will be awake because of sending the outbound packet and will probably remain awake for another millisecond or so to receive reply if RTT from AP to ping target is not high). Or you can test against a device which does not put radio to sleep (a laptop with sleep srates on wireless card disabled).

Thank you for this insight.
Actually I just did a quick test like you suggested and got impressed how fast these portable gets to sleep.
I ran ping from tablet to main router and ping was about 2ms; at the same time I issued a ping from router to tablet and I've got 2ms too. As soon as the ping from tablet ended, the other side went immediately 99/100ms.
Didn't know this about mobile devices; I could have never imagined they "cut" so fast as soon as they finish to make traffic.
Thanks again.