To recap your possible options, you could obtain a higher bandwidth link. As @Giuseppe Larosa earlier described, that would be about your best option for bandwidth (and/or least complexity). The only major disadvantage is it's a single point of failure.
Obtaining maxium advantage of multiple links' bandwidth can run into many snags, a major one, is the one you've described, individual flows usually being limited to using just one link, in your case, it's the worst case as a tunnel is generally seen as a single flow.
However, most multi link technologies don't dynamically select links based on individual current link utilization. For example, dual links usually don't effectively provide twice the bandwidth, they average about a 50% increase.
Technologies such as Cisco's PfR (which can perform dynamic load distribution) or per-packet routing, MLPPP or hardware mux come much closer to obtaining all available bandwidth, but some lose some bandwidth to their own embedded control data (also more impactful as packet size decreases).
Regardless of whether you have a single high bandwidth link multiple lessor bandwidth links, other considerations impact how well you leverage your WAN bandwidth.
As you're tunneling, you should insure you're not fragmenting packets.
Assuming your 50 Mbps links are higher bandwidth Ethernet ports, you should shape to avoid your WAN provider dropping some of your traffic "invisible" to you.
If your shaping shows congestion, you might consider QoS to manage such congestion. (Ideally, if you can, enabling FQ often makes apps work "better" across such congestion.)
