To address your last question first, it's not a vendor-specific problem. If someone wants full-speed, best performance designs for low powered devices like phones and tablets, then they should be looking at a more dense coverage model with APs at lower power. That's true with Cisco, Aruba, HP, Aerohive, Ruckus, etc. That said, you CAN make lower AP density models work with lower powered devices, but it will require more tuning and adjustments to work 100% without issue. Most of that tuning is to add or remove MCS rates, adjust PRT, CSR, Client-Match tuning, etc (in Aruba-speak).
Far too often, you have partners and customers just wanting to do a 1:1 replacement of a legacy 2.4Ghz system (designed around much lower data rates with 2.4Ghz coverage models that supported higher powered laptops in the past), with a 5Ghz VoiP/Video support model for high density client environments filled with low powered phones and tablets. That doesn't work well, and slowly but surely people are figuring that out. It's a physics problem, not a vendor problem (and TxBF is not the cure-all for that issue either as some vendor's proprietary TxBF solutions introduce other problems as well).
-----
To your first point, the same issues still applies, as the link budget still is not symmetrical. However, since you are increasing or magnifying the RF with the directional antenna on the transmit (Tx) side, you are also increasing the Receive (Rx) 'sensitiviy (or gain) on the AP so that it can hear clients farther away. We have gone up to a km distance with 14dB directional antennas to laptops, about 600m to tablets/phones with clear line of sight (which is a pretty long way) and got reliable connections. Ultimately there is a limit with how far most consumer devices can go that has more to do with PHY rate limits than RF limits. But the same premise is true on the asymetrical link budget between a client and AP with high gain directional antenna.
Here's an example. We will use dB since the math is easier. Positive numbers are gain, negative numbers are loss, and these are just values to make a point, actuals would need to be calculated based on a set distance, frequencies used, etc and assumes similar receive sensitivities on each side (AP and Client).
AP to client:
(+20dB radio)+(14dBi antenna)+(-100dB approx free space path loss over 1km)+(2dBi client antenna) = Client's receive SNR of around -64dB
Client to AP:
(+14dB radio)+(2dBi antenna)+(-100dB approx free space path loss over 1km)+(14dBi AP antenna) = AP's receive SNR of around of -70dB
So in the above example, which is JUST a numbers excercise, the actual SNR at 1km would be much lower due to other factors (interference, noise floor, any MRC or lack therof, etc), the AP hears the client at 4x less 'power', or with a 6dB SNR deficiency. Also don't forget that noise floor, interference, etc.
Again, if your client devices are mostly high powered devices and can use external antennas, then you have far more options and can spread out the APs like in the good ol days. But the general concept (again due to physics, not vendor) is that the smaller and more low-powered devices you plan to support, the smaller your AP cell sizes should be to accomodate.
That said, it doesn't mean run your APs at the *exact* power as your lowest powered client devices. But in the OP's example, if the AP's are all wide open at 23dB, and they are having issues with phones and tablets which may max out around 14-15dB, that is a *significiant* link budget difference that can cause some drastic performance issues in regards to roaming, client 'stickyness', throughput, etc.
Hope that helps.