As you may know traffic shaping is essentially traffic delaying in means of buffering excess packets and releasing them in a schedule that shall achieve desired speed restrictions. This implies use of more memory and higher CPU which is an area where SMBs are quite limited. Hence why it is not recommended to enable QoS on these devices.
Low traffic shaping limits usually require large processing buffers. That's why they are not recommended on systems with limited resources.
While we talk about traffic shaping we shall not forget that QoS is also capable of traffic prioritizing in order to guarantee some applications (such as VoIP) the low latency they need to operate properly.
For the outbound traffic, shaping makes sense and shall work even on SMB providing you have set properly interface limits and QoS policy.
For the inbound traffic, shaping makes no sense (except if you are an ISP) because packets have already arrived on the interface and delaying them does not bring any useful benefit. However, traffic prioritizing makes sense here as well. QoS on inbound traffic is good in case of VoIP or other low latency traffic.
It is worth mentioning that as most traffic (broadcast and multicast excluded) sessions are duplex by limiting the outbound traffic you are also limiting the inbound one.
The difference between Guarantee and Limit is apparent - the first one is a weight ratio that limits lower value to not less than XX% (the avail. bandwidth) and no restriction on the upper one. The Limit on the other hand is a restriction on the upper limit and is usually a fixed value (say 1Mbps).
You could try to play with preset interface limits and also disable SecureXL temporarily to see if that will make any difference. If your SMB is constantly bombarded with outbound traffic you could think of offloading traffic shaping job to a more capable device deployed right before the gateway. These devices are just not built for such things. Security and connectivity in one place has its price you know...