We've been incredibly successful with switch-centric designs over the years and have deployed it on every continent except Antarctica. Switch stacks are not a single point of failure as they form a pair of HA switches that are logically a single switch from sa spanning tree perspective.
Using a switch stack at the core of a small data center network provides a number of benefits.
1) In MikroTik routers, which are CPU and not ASIC based, it allows for the traffic to be spread across multiple lanes into the CPU for a single transit subnet and not just one. This is especially important if you plan to run MPLS as there is no ECMP for LDP in RouterOS and you'll be unable to increase capacity by adding links at layer 3.
2) Operational efficiency - by using VLANs to define traffic paths instead of ports, the physical layer is abstracted and changes become far easier and more templated. New services require a new VLAN interface and subnet under the LACP channel.
3) Upgrades or replacement of routers become very quick and easy and less impacting, by putting all of the config on VLANs, very few changes are required to the config to move to another router. Because the switches terminate all physical connections, replacing or upgrading a router doesn't require disruption of cabling to other routers.
4) Network Function Virtualization (NFV) - If you want to connect a hypervisor and run virtual routers, you need a Layer 3 switch stack as Layer 3 MLAG on disaggregated switches is very unreliable. A layer 3 switch stack provides a single Layer point for NFV routers to connect to and form routing adjacencies. This is a critical element in running CHRs successfully.
5) Increase the number of physical connections available. If you run out of ports on the router, your design will not scale successfully. Switches have far more ports than routers and more services and systems can be connected in this design and aren't dependent on the physical number of ports on a router.
When we design for larger ISP or Enterprise networks, we'll often use a hybrid approach and design a Layer 3 MPLS backbone with a non-mpls switch centric stack connected to the MPLS backbone at Layer3. This provides the best of both worlds as resources in a data center can leverage the switch-centric design to deliver local services and use MPLS transport to reach other data centers.