This scenario is true of plenty of smartphones shipping today, but of course it's not something that manufacturers advertise and it's potentially difficult to verify.
One should probably also be concerned about wifi firmware, though smartphone wifi is almost exclusively connected via sdio and not able to directly affect main memory.
The biggest concern in systems where baseband and wifi radios are not-too-deeply integrated is driver bugs where input from those subsystems is overly-trusted or not adequately validated -- of course solid drivers should never trust the hardware, even if not actively malicious, it can be horribly buggy.
I suspect most Tegra-based devices do -- though they introduced a combo apps/model Tegra 4i last year, which likely shares resources.
Generally if it has a standalone apps processor that's provided by a different vendor than the modem it probably does.
Even with unified apps/modem designs, some newer SoCs are designed to provide isolation between the cores, but from a tinfoil hat perspective that requires you to trust the SoC vendor (and perhaps the fab), so if you're paranoid you'd probably avoid any combo designs.
For those who understand why that's important, what do you think about CM11 on a Samsung Galaxy Player (no GSM), using wifi VPN to a cheap phone/hotspot which does have GSM baseband, e.g. Firefox phone? Or two Firefox phones, if Android apps aren't important?
Some protection against malicious firmware/hardware can come from ARM's IOMMU with an open-source Type-1 hypervisor, but these are not mainstream yet.
Whatever the technical merits of Blackphone, their marketing is increasing awareness of mobile security. If they can prove demand for this category of solution, it will increase security audits of all mobile hardware & software stacks.
That I agree and I really hope that it works. But on that note, I don't like the name Blackphone. When I hear "black" I associate it with nefarious activities; and that meaning suggests that only those with criminal purposes need privacy.